fix image token output (#4487)

* fix

* fix

* fix

* add test case

* add test case

* add test case

fastdeploy/engine/common_engine.py

@@ -718,6 +718,8 @@ class EngineSevice:
                             content.outputs.token_ids = token_ids
                             content.outputs.text = delta_text
                             new_contents.append(content)
+                        elif content.finished:
+                            new_contents.append(content)
                         else:
                             llm_logger.warning(
                                 f"current tokens need to accumulate, req_id: {request_id} {content.outputs.token_ids}"

tests/engine/test_decode_token.py

@@ -0,0 +1,51 @@
+import unittest
+from unittest.mock import MagicMock, patch
+
+
+class DummyDataProcessor:
+    def __init__(self):
+        self.decode_status = {}
+
+    def ids2tokens(self, token_ids, req_id):
+        return "", [], None
+
+
+class TestDecodeToken(unittest.TestCase):
+    @patch("fastdeploy.engine.common_engine.EngineSevice.__init__", return_value=None)
+    def setUp(self, mock_init):
+        from fastdeploy.engine.common_engine import EngineSevice
+
+        self.obj = EngineSevice(None)
+        self.obj.data_processor = DummyDataProcessor()
+
+    @patch("fastdeploy.engine.common_engine.envs.FD_ENABLE_RETURN_TEXT", True)
+    def test_decode_token_with_text(self):
+        """测试：env 启用 + 返回非空 delta_text"""
+        self.obj.data_processor.ids2tokens = MagicMock(return_value=("hello", [10, 11, 12, 13], None))
+        self.obj.data_processor.decode_status = {"req_1": (1, 3)}
+
+        delta_text, token_ids = self.obj._decode_token([1, 2, 3], "req_1", is_end=False)
+
+        assert delta_text == "hello"
+        assert token_ids == [11, 12]
+
+    @patch("fastdeploy.engine.common_engine.envs.FD_ENABLE_RETURN_TEXT", True)
+    def test_decode_token_empty_text(self):
+        """测试：env 启用 + 返回空 delta_text"""
+        self.obj.data_processor.ids2tokens = MagicMock(return_value=("", [10, 11, 12], None))
+        self.obj.data_processor.decode_status = {"req_1": (0, 2)}
+
+        delta_text, token_ids = self.obj._decode_token([1, 2], "req_1", is_end=False)
+
+        assert delta_text == ""
+        assert token_ids == []
+
+    @patch("fastdeploy.engine.common_engine.envs.FD_ENABLE_RETURN_TEXT", True)
+    def test_decode_token_with_is_end(self):
+        """测试：is_end=True 时 decode_status 被删除"""
+        self.obj.data_processor.ids2tokens = MagicMock(return_value=("bye", [1, 2, 3, 4], None))
+        self.obj.data_processor.decode_status = {"req_2": (0, 2)}
+
+        delta_text, token_ids = self.obj._decode_token([1, 2, 3], "req_2", is_end=True)
+
+        assert "req_2" not in self.obj.data_processor.decode_status

tests/engine/test_send_tokens.py

@@ -0,0 +1,47 @@
+import time
+from unittest import TestCase
+from unittest.mock import MagicMock, patch
+
+
+class TestZmqSendGeneratedTokens(TestCase):
+    @patch("time.sleep", return_value=None)
+    @patch("fastdeploy.engine.common_engine.EngineSevice.__init__", return_value=None)
+    def setUp(self, mock_init, mock_sleep):
+        from fastdeploy.engine.common_engine import EngineSevice
+
+        self.obj = EngineSevice(None)
+        self.obj.running = True
+
+        # mock 依赖组件
+        self.obj.scheduler = MagicMock()
+        self.obj.send_response_server = MagicMock()
+        self.obj._decode_token = MagicMock()
+        self.obj._decode_token.return_value = ("decoded_text", [101, 102])
+        self.obj.llm_logger = MagicMock()
+
+    def test_zmq_send_generated_tokens_normal_case(self):
+        mock_output = MagicMock()
+        mock_output.outputs.decode_type = 0
+        mock_output.outputs.token_ids = [1, 2, 3]
+        mock_output.finished = True
+
+        self.obj.scheduler.get_results.side_effect = [
+            {"req_1": [mock_output]},
+            {},
+        ]
+
+        def stop_running():
+            time.sleep(0.01)
+            self.obj.running = False
+
+        import threading
+
+        threading.Thread(target=stop_running).start()
+
+        self.obj._zmq_send_generated_tokens()
+
+        self.obj.send_response_server.send_response.assert_called_once()
+        args, kwargs = self.obj.send_response_server.send_response.call_args
+        assert args[0] == "req_1"
+        assert isinstance(args[1], list)
+        assert args[1][0].outputs.text == "decoded_text"